SPCC: Stockfish Contempt-Experiment finished

[pohl4711]

The Stockfish-Contempt-Experiment is finished. Really interesting results IMHO.

Endless RoundRobin-tournament updated, too.

http://spcc.beepworld.de


(Perhaps you have to clear your browsercache or reload the website)

[JJJ]

The Stockfish-Contempt-Experiment is finished. Really interesting results IMHO.

Endless RoundRobin-tournament updated, too.

http://spcc.beepworld.de


(Perhaps you have to clear your browsercache or reload the website)

That's mean Stockfish is now so better than Komodo that he can play with Contempt and make a better score.

[JJJ]

It could also be interesting to experiment with Contempt 75 and Even 100.

[Ozymandias]

Nice experiment. Can you provide average game duration, for each contempt setting?

[pohl4711]

Nice experiment. Can you provide average game duration, for each contempt setting?

Average game-duration of all games with all Contempts was 250 seconds per game(displayed by LittleBlitzerGUI). I cannot say, how much longer (moves or time) the games with higher Contempt were, but I can say, that they were a bit longer.
If you want more information, contact me on my website and I will send the 5000 games to your email. I am very busy in the next days, so I cant do more analysis on this database by my own.

Stefan

[Vinvin]

http://spcc.beepworld.de/experiments.htm

C=+25 - 580.5/1000
C=+35 - 565.0/1000
C=+50 - 579.0/1000

Probably not enough games to have reliable numbers
I did find no logical reason to justify +0.35 is worse than +0.25 and +0.50 ...

[JJJ]

http://spcc.beepworld.de/experiments.htm

C=+25 580.5/1000
C=+35 565.0/1000
C=+50 579.0/1000

Probably not enough games to have reliable numbers
I did find no logical reason to justify +0.35 is worse than +0.25 and +0.50 ...

Me too. But we can notice than almost any contempt improved the score.

[Ozymandias]

Average game-duration of all games with all Contempts was 250 seconds per game(displayed by LittleBlitzerGUI).

If you want more information, contact me on my website and I will send the 5000 games to your email.

I was interested in seeing the correlation, between 3folds and number of moves, but if you're using LittleBlitzer, reconstructing the full games would be a lot of work.